GPU speed sifferences

[BVA]

PLEASE can someone explain to me what can cause the speed difference?
I have used 2 identical machines, (different motherboards but otherwise almost identical. One have a gtx 980Ti and the other have a 1080.

The 1080 rendered 8m2.3s and the 980Ti completed in 3m53s

That is less than half the time on an older card? I have also picked up that denoising is a lot slower on the 1080 than the 980Ti.

Both are on 372.70 nvidia driver, both on windows 10.

[BVA]

I have just run the arion benchmark on both machines. the 1080 is faster than the 980Ti with that benchmark

Can it be Vray does not support the new card properly yet?

[vladimir.nedev]

The screen-shots you posted have a different resolution.
This could affect render time quite a bit.

Greetings,
Vladimir Nedev

[BVA]

oops...maybe the auto resize was ticked without me noticing it. I will run it again, thanks for pointing out.

[BVA]

no, I made sure and the 980Ti is still faster than the 1080

[vladimir.nedev]

That's interesting, is there a monitor attached to one of them ?

We have some additional statistics about GPU utilization, that I need to show in the top right corner of the image.
They might help in understanding what's going on.

Greetings,
Vladimir Nedev

[MildMustard]

Extra figures. Attached is rendered from the same file as Barend's. Notably my Primary and secondary Pascal cards have a 26s difference. Previous test I did the Primary card was faster and the secondary was slower, so it isn't always consistent. PCI-e lanes for both Pascal cards are x16 and the Maxwell x8. I know SLI doesn't play a roll in GPU rendering but is there a disadvantage of having the SLI connector on (for the Pascal cards)?

[BVA]

That's interesting, is there a monitor attached to one of them ?

I was actually wondering if a monitor can make this big difference?

the 1080 have 2x monitors attached.
1. 3440x1440
2. 1920x1080

the 980Ti have 2 monitors attached
both at 1920x1080 resolution

[GidPDX]

Why don't you try running your higher res monitor at 1920x1440 to see?

Also you don't have any other applications open using the GPU right? Like even PS etc.

Still I doubt those factors explain a nearly 40%+ difference. That scene should easily fit in even 2GB of vram AFAIK. 1080 should be faster or at least the same speed from everything we've read/heard! Is it one of those "wait for CUDA 8" things?

[BVA]

Yes...something is not working. I will do some more tests... I will disconnect extra monitors and just use the same resolution etc.
No other software run while I tested these

[savage309]

Many things could matter - it could be a good idea to turn SLI off, in my tests a while ago having monitor had ~10% performance loss, also a clocked 980ti can be faster than reference (founders edition) 1080.

Best,
Blago.

[BVA]

I have run the tests again, both pc's with one monitor at the same resolution.

the 980 completed in 5m51 actually went a little slower (10 sec), not sure why.
the 1080 completed in 7m33sec. (roughly 30sec faster)

both screen cards were just default drivers installed, no OC.

on the 1080 I tried different profiles that came with the card (Asus, GPU tweak). even on the OC profile it renders 8m5 sec......faster than the original but still a lot slower than the 980Ti.

This is the two GPU's I compare.
the 1080 :https://www.asus.com/us/Graphics-Car...BO-GTX1080-8G/
980Ti: http://www.palit.com/palit/vgapro.ph...B-2000J&tab=ov

Please can someone explain to me what can cause the difference? I am new to GPU. I am considering getting 2 more cards to push GPU a little more. I was planning on adding 2x 1080's so I have 3x 1080's in one machine....now I am not sure it is the best option?

[MildMustard]

I wonder if it is the cuda core count?...

your 980 Ti has 2816 (mis-print?)
your 1080 has 2560

I thought faster clock speed compensates but doesn't seem so. Could also be we need to "wait for cuda8/optimisations"...

[vladimir.nedev]

Can you try increasing the GPU Ray Bundle Size to 1024 and GPU Rays per pixel to 16, then run the tests again ?

It's possible that the result is scene dependent, so trying another scene with a different composition of lights and materials might be a good idea.

Greetings,
Vladimir Nedev

[GidPDX]

I wonder if it is the cuda core count?...

your 980 Ti has 2816 (mis-print?)
your 1080 has 2560

I thought faster clock speed compensates but doesn't seem so. Could also be we need to "wait for cuda8/optimisations"...

Don't go by CUDA core count- things are more complicated than that and would not explain the speed difference of a 980ti and a 1080 GTX that BVA is seeing.

Many benchmarks point to the 1070 even being potentially faster than the 980ti with CUDA applications- and it has fewer cores:

http://barefeats.com/geekbench4.html
https://www.phoronix.com/scan.php?pa...New-Benchmarks

I suspect though that Vray isn't optimized yet for the 1080, things will change as you've said with CUDA 8, or there is something unique to BVA's over all build etc. Cuda 8 was just released 2 days ago, so not sure CG has really had time to optimize things yet.

BVA- have you disabled SLI yet? I don't know about Vray, but for Octane it's highly recommended: "No, but it can use multiple video cards for rendering (see above). It is also recommended to disable the SLI option in your NVIDIA control panel to maximize Octane’s rendering performance." Otoy faq

It also stands that if you haven't, as simple trite as it is, try downloading the latest drivers for the card, un-installing the old, and installing the new ones. Several people have reported that it helps with some issues- not sure if this is the case.

EDIT- well it turns out that from what Blago says the pascal lineup should be fully optimized for RT. That squashes the theory that they need CUDA 8 to get RT optimized for the 1080 GTX.

To help get a better fix on where the issue is you should probably also test some of the same benchmark scenes that CG used here if you can get your hands on them:

https://plus.google.com/+VladimirKoy...ts/R1XPacoWvB4